Author: Robert Hyatt
Date: 12:10:59 10/28/98
Go up one level in this thread
On October 28, 1998 at 08:19:39, Ernst A. Heinz wrote: >On October 27, 1998 at 22:00:54, Robert Hyatt wrote: >> >> [...] >> >>Simple. I do a *deep* search, and only do the pawn race evaluations at the >>*tips* and not on interior nodes. I depend on the search to find the mates, >>the ways the opponent can create an even quicker-queening passer... and only >>after the normal 20+ plies of search are exhausted do I do the "race" test. > >Delegating the race evaluation completely to the search is indeed simple but >actually far from satisfying (i.e., it is *too* simple IMHO). > >Of course, if you have already reached an K+P endgame you can easily do the 20+ >plies of search. Yet, it is impossible to resolve Pawn races by pure search if >you are on the verge of entering the according endgame. I know that all >quiescence searches are full of errors but still I do not want my evaluation to >return a premature winning score only because one side features an unstoppable >passed Pawn which is not worth much in reality. I think it is important to do >more than that in the evaluation in order to enable the search to find better >ways for *trading into* K+P endgames. > >=Ernst= I don't delegate the "race" to the search... I delegate all the other stuff. I have a piece of code that very accurately says that one pawn queens before the other... and it understands tempo, queening with check, queening and attacking the other pawn's queening square (if the king isn't defending that square) and so forth. But all it says is "if both sides push pawns" this side wins or they both promote. I let the search handle the cases where pawn pushes are not the only thing going on... ie checks, 3 pawns + king bearing on the opponent king, etc... But since I don't trust a quiescence search to find those things anyway, this works perfectly well. I haven't lost a won/drawn game, nor drawn a won ending because of this that I can recall... The really hard cases are those few where you push but your opponent can do something else, like threaten mate. Those would cause problems...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.